{
  "id": 205799,
  "title": "How to combine models ? ",
  "url": "/competitions/riiid-test-answer-prediction/discussion/205799",
  "author_name": "",
  "post_date": "2020-12-21T22:20:22.911297600Z",
  "votes": null,
  "comment_count": 3,
  "views": 0,
  "content": "<p>PS: This is my first kaggle competition<br>\nI noticed that usually in Kaggle competition, combining models is what usually wins the competition. In the context of this competition, how could we combine two model? </p>",
  "messages": [
    {
      "id": "1121765",
      "postDate": "12/21/2020 22:20:22",
      "content": "<p>PS: This is my first kaggle competition<br>\nI noticed that usually in Kaggle competition, combining models is what usually wins the competition. In the context of this competition, how could we combine two model? </p>",
      "rawMarkdown": "PS: This is my first kaggle competition\nI noticed that usually in Kaggle competition, combining models is what usually wins the competition. In the context of this competition, how could we combine two model?",
      "votes": null
    },
    {
      "id": "1121983",
      "postDate": "12/22/2020 04:58:18",
      "content": "<p>In this competition, there are some public notebooks about combining models (ensemble), so you can learn how to do that from them.</p>",
      "rawMarkdown": "In this competition, there are some public notebooks about combining models (ensemble), so you can learn how to do that from them.",
      "votes": null
    },
    {
      "id": "1122228",
      "postDate": "12/22/2020 09:17:34",
      "content": "<p>Hi Abdessalem, I'm no expert but I've learned know some methods to effectively combine models. The simplest approach is to average the output of two or more models. The rule of thumb is to average the output of dissimilar models. The idea here is that if two models make the same kind of prediction mistake, averaging doesn't help, but if they make different mistakes, averaging mitigates the errors made.</p>\n<p>The other way to combine models is by stacking, where the output of one model is the input for another. You create a model to output the predictor, which you then use as an input to your main model.</p>\n<p>I hope this helps!</p>",
      "rawMarkdown": "Hi Abdessalem, I'm no expert but I've learned know some methods to effectively combine models. The simplest approach is to average the output of two or more models. The rule of thumb is to average the output of dissimilar models. The idea here is that if two models make the same kind of prediction mistake, averaging doesn't help, but if they make different mistakes, averaging mitigates the errors made.\n\nThe other way to combine models is by stacking, where the output of one model is the input for another. You create a model to output the predictor, which you then use as an input to your main model.\n\nI hope this helps!",
      "votes": null
    },
    {
      "id": "1122235",
      "postDate": "12/22/2020 09:27:49",
      "content": "<p>Thanks, this is indeed helpful.</p>",
      "rawMarkdown": "Thanks, this is indeed helpful.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1121983,
      "author_name": "japandata509",
      "author_url": "",
      "post_date": "12/22/2020 04:58:18",
      "content": "<p>In this competition, there are some public notebooks about combining models (ensemble), so you can learn how to do that from them.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1122228,
      "author_name": "richardcsuwandi",
      "author_url": "",
      "post_date": "12/22/2020 09:17:34",
      "content": "<p>Hi Abdessalem, I'm no expert but I've learned know some methods to effectively combine models. The simplest approach is to average the output of two or more models. The rule of thumb is to average the output of dissimilar models. The idea here is that if two models make the same kind of prediction mistake, averaging doesn't help, but if they make different mistakes, averaging mitigates the errors made.</p>\n<p>The other way to combine models is by stacking, where the output of one model is the input for another. You create a model to output the predictor, which you then use as an input to your main model.</p>\n<p>I hope this helps!</p>",
      "votes": null,
      "replies": [
        {
          "id": 1122235,
          "author_name": "abdessalemboukil",
          "author_url": "",
          "post_date": "12/22/2020 09:27:49",
          "content": "<p>Thanks, this is indeed helpful.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1121765": "PS: This is my first kaggle competition\nI noticed that usually in Kaggle competition, combining models is what usually wins the competition. In the context of this competition, how could we combine two model?",
    "1121983": "In this competition, there are some public notebooks about combining models (ensemble), so you can learn how to do that from them.",
    "1122228": "Hi Abdessalem, I'm no expert but I've learned know some methods to effectively combine models. The simplest approach is to average the output of two or more models. The rule of thumb is to average the output of dissimilar models. The idea here is that if two models make the same kind of prediction mistake, averaging doesn't help, but if they make different mistakes, averaging mitigates the errors made.\n\nThe other way to combine models is by stacking, where the output of one model is the input for another. You create a model to output the predictor, which you then use as an input to your main model.\n\nI hope this helps!",
    "1122235": "Thanks, this is indeed helpful."
  },
  "source": "meta"
}